Design Thinking with AI: what changes and what doesn't

I gave a Design Thinking class last semester at UDD's School of Innovation in Santiago. The assignment was simple: go out, observe real users, come back with what you learned.

Twenty-four hours later, half the class returned with beautifully organized insights — synthesized, categorized, visualized in clean decks. The other half came back with notebooks full of handwritten observations, stories, and contradictions they didn't quite know what to do with.

The first group had used AI.

The second group had talked to people.

When I asked the AI group to tell me about a specific user — what their day actually looked like, what frustrated them, what they hoped for — most of them couldn't. They had the pattern. They didn't have the person.

Where AI genuinely helps

Design Thinking — as formalized by IDEO and diffused globally through Stanford's d.school (Brown, 2009) — is a five-stage iterative process: Empathize, Define, Ideate, Prototype, Test. A corrective to the bias of designing for yourself. A method for placing the user at the center.

In four of these five stages, AI is a genuine accelerant.

These are real gains. Herbert Simon understood that the fundamental mechanism of design intelligence is iteration (1969) — testing and adjusting until something works. AI dramatically reduces the cost of each iteration. A team that can test three ideas in the time it previously took to test one is genuinely better positioned.

The one stage that resists

Empathy is different in kind, not just degree.

In the Design Thinking sense, empathy is not sentiment. It's the deliberate, effortful practice of suspending your own assumptions long enough to genuinely encounter someone else's experience. It requires physical presence, productive discomfort, and — critically — the willingness to be wrong about what you expected to find.

AI can summarize what people said. It cannot replicate the experience of watching someone struggle with something they've accepted as normal, and feeling the specific texture of that frustration. That moment — when the designer perceives something the user has stopped seeing — is where genuine insight lives.

It is not transferable to a prompt.

Tim Brown called this the difference between technical empathy and creative empathy (Brown, 2009). Technical empathy aggregates. Creative empathy understands. The first is pattern recognition. The second is meaning-making. AI executes the former with exceptional efficiency. The latter remains irreducibly human — for now, and probably for longer than we'd like to believe.

The new risk

The risk AI introduces into Design Thinking is not that it replaces the process.

It's that it makes a shortcut feel like the real thing.

Students who use AI in the Empathy stage often produce outputs that look rigorous. The categorization is clean. The language is precise. The insights are plausible. What's missing is the friction of actual contact — the unexpected detail, the thing that fits no category. That friction is not noise. It is signal.

Dweck's research on mindset suggests that the willingness to engage with difficulty — rather than route around it — is what produces genuine insight (2006). The IDEO concept of creative confidence isn't the confidence to use tools — it's the confidence to be uncertain in front of a real problem and trust the process (Kelley & Kelley, 2013). Over-reliance on AI in the early stages quietly erodes this confidence by making uncertainty feel unnecessary.

Uncertainty is the raw material. The process is designed to work with it, not around it.

The new job

The designer's question has shifted. It's no longer "can I generate insights?" AI can do that. It's "can I tell the difference between an AI-generated plausibility and a genuine human truth?"

That distinction requires having done the fieldwork yourself. At least enough to know what real contact with a problem feels like. Without that reference point, you have no way to evaluate what the AI is giving you.

You need to have been wrong about a user before you can recognize when the AI is being wrong about one.

AI makes you faster at everything. The question is whether you're going faster in the right direction.

The only way to know the direction is to have looked at it yourself first.